I think I have a handle on what is happending... The modeline redisplay code calls generate_formatted_string_db which does . . generate_fstring_runes (w, &data, 0, 0, -1, format_str, 0, max_pixpos - min_pixpos, findex, type, &offset, Qnil); [This calls add_glyph deep inside. Instead of only one element to the display block it now add all the characters of the string making it much longer than with the old text-glyph code] resize_string (XSTRING (result_str), -1, data.bytepos - XSTRING_LENGTH (result_str)); [Reserve space] for (elt = 0, len = 0; elt < Dynarr_length (db->runes); elt++) { if (Dynarr_atp (db->runes, elt)->type == RUNE_CHAR) { len += (set_charptr_emchar (strdata + len, Dynarr_atp (db->runes, elt)->object.chr.ch)); } } [copies the chars from the display block] What probably happens is that with the new text glyph code there are more RUNE_CHAR objects than excpected and we blindely overwrite wahtever string came after the modeline string... Sounds plausible? Jan P.S. The reason we probably see this with display-time is that the X first, tty later route causes the full path of the XPM file to be put in as a really long text glyph.